spectrum sharing at 3.5 ghz - nist
TRANSCRIPT
Spectrum Sharing at 3.5 GHzContributors: Raied Caromi, Paul Hale, Tim Hall, Thao Nguyen, Anirudha Sahoo, Michael Souryal, and Adam Wunderlich
Outline
Contributions to 3.5 GHz Citizens Broadband Radio Service (CBRS)
• Release 1 of Wireless Innovation Forum CBRS standards– NIST-led development of federal incumbent protection test procedures
• NIST reference model for federal incumbent protection– Used by regulator to test and certify commercial Spectrum Access Systems
Detection of Incumbent Signals in the 3.5 GHz Band
• Characterization of incumbent signals– 3.5 GHz radar waveform measurements (NASCTN)
• Machine-learning radar detection and classification algorithms
• RF signal data sets
• Sensor placement and configuration
2
Citizens Broadband Radio Service Standards and Reference Model
Citizens Broadband Radio Service (CBRS)
Opens 150 MHz of prime spectrum to traditional mobile operators and new entrants; shared access to 3.5 GHz band (3550 MHz to 3700 MHz)
• Jan 2015: CBRS standards process begins under the auspices of the Wireless Innovation Forum (WINNF).
• April 2015: FCC finalizes CBRS rules.
• Jan 2016: NIST joins the WINNF as a full member.
• Jan 2018: Completion of Release 1specifications
4
CBRS Architecture
Three Tiers of Spectrum Users
1. Incumbent
– Federal (e.g., U.S. Navy radars)
– Fixed satellite service earth stations
– Grandfathered wireless broadband licensees
2. Priority access licensee (PAL)
3. General authorized access (GAA)
Source: https://www.cbrsalliance.org/blog
5
CBRS ArchitectureCBRS devices (CBSDs): fixed base stations/access points operating as PALs and GAAs (typically LTE cells) providing mobile broadband service to end user devices
Spectrum Access System (SAS): allocates spectrum resources (frequencies, power) to CBSDs; manages interference of lower tiers to higher tiers and among Tier 2
Environmental Sensing Capability (ESC): sensor network that monitors CBRS band for federal incumbent signals (military radar); upon detecting an incumbent in a Dynamic Protection Area (DPA), SAS reconfigures CBSDs to mitigate interference in the DPA within 300 s
Source: https://www.cbrsalliance.org/blog
Federal Incumbent
CBSDs
ESC sensor
6
Federal Incumbent Dynamic Protection Areas (DPAs)
East CoastWest CoastGulf CoastAlaskaHawaiiPuerto RicoGuamInland sites
USS George H.W. Bush. Credit: U.S. Navy photo by Mass Communication Specialist 3rd Class Nicholas Hall [Public domain], via Wikimedia Commons
7
52% of the U.S. population lives in coastal watershed counties [NOAA].
NIST Contributes to CBRS Standards
• WINNF-TS-0112, CBRS Operational and Functional Requirements
• WINNF-TS-0016, SAS to CBSD Protocol Specification
• WINNF-TS-0061, SAS Test and Certification Specification
Part of suite of 10 standards completed 2018https://cbrs.wirelessinnovation.org/release-1-standards-specifications
8
NIST Contributes to Spectrum Access System (SAS) Testing
NIST/CTL
Created the federal incumbent (DPA) protection reference model
NTIA
Performed certification testing of commercial SASs
Wireless Innovation Forum
Published test and certification
specifications
SAS Industry
Developed the SAS Test Harness
CBSD Emulator
SAS Emulator
ESC Emulator
Propagation
Interference
PAL Protection
DPA Protection
Reference Models:
SAS Test Harness
SAS Under Test
SAS-CBSD Interface
SAS-SAS Interface
SAS-ESC Interface
https://github.com/Wireless-Innovation-Forum/ 9
Federal Incumbent Protection Procedure• Sensor detects offshore shipborne radar
and informs SAS of activated DPA and channel
• SAS reallocates spectrum resources to mitigate interference to active DPA
– Uses a pre-calculated “move list” of CBSDs
– Moves CBSDs off the protected channel in the vicinity of the DPA
• Move list is designed to meet the required interference protection level:
95th percentile of aggregate interference into radar receiver antenna –144 dBm/10 MHz
at every point in the DPADoD-defined protection level
Dynamic Protection Area (DPA)
Radar sensor
CBRS cell station
10
Move-List Example: New York Offshore DPA
• Simulated deployment of 80K CBSDs (generated with NTIA/OSM software)
• “Move list” calculated by NIST reference model of DPA protection algorithm
– Using ITM reference propagation model
– Based on 2000 Monte Carlo realizations of each path (modeling time variability of path loss)
Google EarthImage Landsat/Copernicus© 2018 GoogleImage NOAAData SIO, NOAA, U.S. Navy, NGA, GEBCO
11
CBSDs:red: on move listblue: not on move listwith dot: Category B
NIST Brings Uncertainty Analysis to SAS Testing
• Federal incumbent protection test procedure includes an interference margin to account for move-list uncertainty.
• NIST proposed a methodology adopted by the WINNF for quantifying this margin.
Example: San Diego DPA (“West 14”)
• Simulated deployment of 22K CBSDs.
• Generated 100 independent move lists.
• 1.5 dB margin relative to protection level is sufficient in this case.
12
Industry Statements Following CBRS Standards Release
“… it is now possible to make major changes in how we approach managing spectrum resources …… flexible, cloud-based management can … relegate the exclusive, command and control vision of spectrum allocation to history.” Eric Schmidt, Executive Chairman of Google parent company Alphabet Inc.
“… 150MHz of spectrum in a continuous block. This spectrum will enable cost effective coverage and capacity expansion at large scale.” Ricky Corker, Executive Vice President, Nokia
“Ericsson’s commitment to supplying LTE equipment for the band will ensure strong commercial support for the ecosystem.” Paul Challoner, VP Network Product Solutions, Ericsson Inc., USA
“We look forward to offering innovative new products and services in this new shared CBRS spectrum.” Ed Chan, Chief Technology Architect and Network Planning, Verizon 13
Industry Statements Following CBRS Standards Release
“The efforts of our company, NIST and the other members of the WINNF SSC to establish standards, testing and certification for spectrum sharing are setting the stage for improving wireless service indoors, expanding broadband services to rural areas, and providing private wireless capabilities for industrial users. It’s an outstanding example of public-private collaboration,” Kurt Schaubach, Chief Technology Officer, Federated Wireless
14
Detection of RF Signals in Shared Spectrum
16
Disseminating curated RF signal waveforms for training AI models, derived from both measurements and synthesis
Data Sets
Detection of RF Signals in Shared Spectrum
Leveraging state-of-the-art machine learning models, including deep-learning
neural networks
Improved Detectors
Algorithms for efficient placement and configuration
of sensors to detect incumbent signals
Sensor Deployment
• Top of small building• ~20 m elevation• Trees and small buildings to
the west
NASCTN 3.5 GHz Radar Waveform Measurements
Objectives• Characterize shipborne SPN-43 radar• Create waveform library
Measurements collected• Two naval bases: San Diego, Norfolk• 2 months at each location
Outcomes• 14,739 spectrograms, every 10 min• 3,336 sixty-second IQ waveform
captures (200 MHz inst. bandwidth)
17
3.5 GHz Spectrograms
from NASCTN Report 2, NIST Technical Note 1954, “3.5 GHz Radar Waveform Capture at Point Loma: Final Test Report”
SPN-43 signal without adjacent-band radar emissions SPN-43 signal embedded in adjacent-band radar emissions
18
ESC sensors are supposed to detect the SPN-43 signal in both cases.
Classifying and Labeling 3.5 GHz Spectrograms
• Evaluated 13 algorithms for classifying the spectrograms for SPN-43 presence
– 8 deep learning architectures
– 3 classical machine learning methods
– 2 classical signal detection methods
• Found that 3-layer convolutional neural network (CNN) offered superior tradeoff between accuracy and computational complexity
• Trained CNN used to label the complete set of spectrograms for SPN-43 presence and derive channel occupancy and ambient power statistics
Publications:
W. M. Lees, A. Wunderlich, P. Jeavons, P. Hale, and M. Souryal, “Deep learning classification of 3.5 GHz band spectrograms with applications to spectrum sensing,” IEEE Transactions on Cognitive Communications and Networking, June 2019.
W. M. Lees, A. Wunderlich, P. Jeavons, P. Hale, M. Souryal, “Spectrum Occupancy and Ambient Power Distributions for the 3.5 GHz Band Estimated from Observations at Point Loma and Fort Story,” NIST Technical Note 2016, Sept 2018.
19
Type Occupancy Ratio
Median 75th Pct 99th Pct
Occupancy 0.287±0.005 20 min 90 min 3630 min
Vacancy 70 min 230 min 1450 min
Classifier Performance and Band StatisticsAmong 13 algorithms tested, best in each class were:
• Convolutional neural network with three layers (CNN-3)
• Inception, a deep neural network developed for computer vision
• Support vector machine (SVM)
• Energy detection correlated with radar antenna sweep period (SI-ED)
CNN-labelled spectrograms used to derive statistics of CBRS channel occupancy, vacancy, and noise-plus-interference levels
20
3.5 GHz High-Resolution Waveforms60-second captures
4 ns sampling interval
21
Test Waveform Generation
Field-measured
radar signals
Test Waveform Generator
Test waveforms
Signal Generator
Field-measured
interference signals
Synthetic interference
signals and/or WGN
Radar detection algorithm simulation
RF sensor under test
22
(e.g., 3.5 GHz ESC sensor)
Test Waveform ExamplesSpectrogram of in-band radar signal with co-channel LTE Time-domain plot of a 0.2s portion of the in-band radar
signal with co-channel LTE (at the 203s mark of the test)
23R. Caromi et al., “3.5 GHz ESC Sensor Test Apparatus Using Field-Measured Waveforms,” in Proc. Wireless Innovation Forum Summit on Wireless Communications Technologies (WInnComm), Nov. 2018.
SPN-43 shipborne radar Radar emissions
from adjacent band
dBm
Radar Detection with Full IQ Waveforms
Radar sensor
CBRS cell station
Motivation: Commercial-federal spectrum sharing relies on a network of commercial sensors that detect federal incumbent signals in presence of commercial signals and out-of-band emissions from adjacent bands.
Question: How well can sensors detect incumbent signals in the band?
Classical Approach: Filter matched to known parameters
Problem: Sensors may have only partial knowledge of incumbent waveforms and interference.
New Approach: Machine learning model trained to identify incumbent signals embedded in noise/interference
24
NIST is Applying Machine-Learning to RF Detection
Feature-based support-vector machines• Higher-order statistics classifier• Peak-analysis classifier
Less computationally expensive to generalize than a matched filter (conventional)
Evaluated with field measurements of 3.5 GHz band (+ simulated LTE)
Peak-analysis classifier excels at distinguishing in-band from adjacent-band radars.
Seeing promising results from deep learning neural networks
Adjacent-Band Radar Interference at 10 dB to 20 dB interference-to-
noise ratio
LTE Interference at –102 dBm/MHz
25
RF Signal Data Sets for Training AI Models
Recognized data sets exist in domains such as speech, object, and handwriting recognition.
“No robust competition data sets exist in the ... radio domain.”*
* T. J. O’Shea, J. Corgan, “Convolutional radio modulation recognition networks,” CoRR, 2016.
MNIST dataset [Josef Steppan under CC BY-SA 4.0]
26
RF Signal Data: Characteristics and Sources
Goal: Publicly Available Database of RF Signal Data• Targets next generation
communication systems• Composed of traceable data• Curated to produce a representative
range of signals, noise, and channels
Sources• Field measurements• Testbeds• Simulation• Other organizations NIST/CTL Surrogate Radar Waveform Generator
27
Deployment of RF Sensors for Incumbent Detection
Main Contribution
• Methodology for determining placement and antenna configuration of sensors to detect incumbent operations, using:
– Existing tower sites as candidate locations for the sensors
– Irregular Terrain Model (ITM) in point-to-point mode for path loss
Applications
• Planning and deployment of sensors by commercial operators (e.g., 3.5 GHz ESC)
• Evaluation and certification of sensor network designs by regulators
28
Deployment of RF Sensors: Technical Approach
Motivation – CBRS use case:Environmental Sensing Capability (ESC) sensors deployed along the coasts are required to detect the presence of incumbent shipborne radar inside a Dynamic Protection Area (DPA).
Definitions:
Detection coverage – area of DPA covered by sensor(s)
Excess area – area outside a DPA (neighboring DPAs, sea, land) detected by sensor(s)
Objective: Determine locations and parameters of sensor antennas such that:
1. Detection coverage is maximized (protect incumbent)
2. Excess area is minimized (avoid unnecessary disruption of commercial operations)
Sensor Antenna Coverage
29
Sensor Deployment Along U.S. CoastsKey Findings:• A single antenna can fully cover each DPA.• Most sensors employ:
– A maximum antenna height of 100 m– Antennas with small beamwidths and high
gains, e.g., (30°, 18 dBi) and (45°, 17 dBi)• Probability of false alarm:
West Coast East/Gulf CoastsEast Coast DPA
From Neighboring DPAs
Out at SeaTrad
eo
ff
30
NIST Contributions to 3.5 GHz Spectrum Sharing
✓Requirements for SAS and ESC systems (in WINNF-TS-0112, CBRS Operational and Functional Requirements)
✓Procedures for testing CBRS protocols and interference management (in WINNF-TS-0061, SAS Test and Certification Specification)➢ Federal incumbent interference protection➢ Multiconstraint interference protection
✓Reference implementation of the CBRS federal incumbent protection algorithm (https://github.com/Wireless-Innovation-Forum/)
✓CBRS incumbent occupancy and noise statistics from CNN-labelled measurements
✓RF sensor test waveforms comprised of field measurements and generated signals
✓ML models for detection of radar signals embedded in LTE and OOB emissions
✓RF sensor deployment algorithms for efficient coverage of protection areas
✓Technical expertise on protocols, propagation models, and interference models to standards groups and to the Federal 3.5 GHz Joint Working Group
31
Recent Publications and Patents1. "Deep Learning Classification of 3.5 GHz Band Spectrograms with Applications to Spectrum Sensing," W. M. Lees, A.
Wunderlich, P. Jeavons, P. D. Hale, and M. R. Souryal, IEEE Transactions on Cognitive Communications and Networking, vol. 5, no. 2, pp. 224-236, June 2019.
2. “Detection of Incumbent Radar in the 3.5 GHz CBRS Band Using Support Vector Machines,” R. Caromi and M. Souryal, in Proc. IEEE Sensor Signal Processing for Defence (SSPD), May 2019.
3. “Optimal Dynamic Spectrum Access Scheme for Utilizing White Space in LTE Systems,” A. Sahoo, T. A. Hall, C. Hagwood, in Proc. IEEE Wireless Communications and Networking Conference (WCNC), April 2019.
4. “A Practical Approach to Place Coastal Sensors for Spectrum Sharing in the 3.5 GHz Band,” A. Sahoo, T. T. Nguyen and T. A. Hall, in Proc. IEEE WCNC, April 2019.
5. “Analytical Modeling of White Space Utilization for a Dynamic Spectrum Access System,” C. Hagwood, A. Sahoo, T. A. Hall, in Proc. IEEE WCNC, April 2019.
6. “Detection of Incumbent Radar in the 3.5 GHz Band,” R. Caromi, M. Souryal, and W.-B. Yang, in Proc. IEEE Global Conference on Signal and Information Processing (GlobalSIP), Nov. 2018.
7. “3.5 GHz ESC Sensor Test Apparatus Using Field-Measured Waveforms,” R. Caromi, J. Mink, C. Wedderburn, M. Souryal, and N. El Ouni, in Proc. Wireless Innovation Forum Summit on Wireless Communications Technologies (WInnComm), Nov. 2018.
8. U. S. Patent Application Pub. No. US 2018/0302920, “Apparatus and Method for Dynamically Controlling Spectrum Access,” T. A. Hall, A. Sahoo, C. Hagwood, S. Streett, Oct. 2018.
9. “3.5 GHz Federal Incumbent Protection Algorithms,” M. R. Souryal, T. T. Nguyen, and N. J. LaSorte, in Proc. IEEE International Symposium on Dynamic Spectrum Access Networks (DySPAN), Oct. 2018.
10. “Sensor Placement and Detection Coverage for Spectrum Sharing in the 3.5 GHz Band,” A. Sahoo, M. Ranganathan, T. Nguyen, and T. Hall, in Proc. IEEE PIMRC, Sept. 2018. 32
Recent Publications and Patents (cont.)11. “Spectrum Occupancy and Ambient Power Distributions for the 3.5 GHz Band Estimated from Observations at Point Loma
and Fort Story,” W. M. Lees, A. Wunderlich, P. Jeavons, P. Hale, and M. Souryal, NIST Technical Note 2016, Sept. 2018.
12. “Dynamic Spectrum Access Algorithms Based on Survival Analysis,” T. Hall, A. Sahoo, C. Hagwood, and S. Streett, IEEE Transactions on Cognitive Communications and Networking, vol. 3, no. 4, pp. 740-751, Dec. 2017.
13. "3.5 GHz Environmental Sensing Capability Detection Thresholds and Deployment," T. T. Nguyen, M. R. Souryal, A. Sahoo, and T. A. Hall, IEEE Transactions on Cognitive Communications and Networking, vol. 3, no. 3, pp. 437-449, Sept. 2017.
14. “3.5 GHz Radar Waveform Capture at Fort Story: Final Test Report,” P. Hale, J. Jargon, P. Jeavons, M. Souryal, A. Wunderlich, and M. Lofquist, NIST Technical Note 1967, NASCTN Report 3, July 2017.
15. “3.5 GHz Radar Waveform Capture at Point Loma: Final Test Report,” P. Hale, J. Jargon, P. Jeavons, M. Souryal, A. Wunderlich, and M. Lofquist, NIST Technical Note 1954, NASCTN Report 2, May 2017.
16. “Exploiting LTE White Space Using Dynamic Spectrum Access Algorithms based on Survival Analysis,” T. Hall, A. Sahoo, C. Hagwood, and S. Streett, in Proc. IEEE International Conference on Communications (ICC), May 2017.
17. “3.5 GHz Environmental Sensing Capability Sensitivity Requirements and Deployment,” T. Nguyen, A. Sahoo, M. Souryal, and T. A. Hall, in Proc. IEEE DySPAN, Mar. 2017.
18. “Fair Resource Allocation in the Citizen Broadband Radio Service Band,” A. Sahoo, in Proc. IEEE DySPAN, Mar. 2017.
19. “An Analytical Model for Inference Attacks on the Incumbent's Frequency in Spectrum Sharing,” A. Ben Mosbah, T. A. Hall, M. Souryal, and H. Afifi, Best Poster Award, in Proc. IEEE DySPAN, Mar. 2017.
20. “Analysis of the Vulnerability of the Incumbent Frequency to Inference Attacks in Spectrum Sharing,” A. Ben Mosbah, T. A. Hall, M. Souryal, and H. Afifi, in Proc. IEEE Consumer Communications & Networking Conference (CCNC), Jan. 2017.
33
Standards Contributions1. “Requirements for Commercial Operation in the U.S. 3550-3700 MHz Citizens Broadband Radio Service Band,” WINNF-TS-
0112, Version V1.7.0, May 7, 2019.2. “Test and Certification for Citizens Broadband Radio Service (CBRS); Conformance and Performance Test Technical
Specification; SAS as Unit Under Test (UUT),” WINNF-TS-0061, Version V1.5.0, Apr. 29, 2019.3. “Signaling Protocols and Procedures for Citizens Broadband Radio Service (CBRS): Spectrum Access System (SAS) - Citizens
Broadband Radio Service Device (CBSD) Interface Technical Specification,” WINNF-TS-0016, Version V.1.2.3, Oct. 31, 2018.4. “Revisions to Federal Incumbent Protection Tests for Inland DPAs,” M. Souryal, WINNF-18-I-00101, WG4, May 22, 2018.5. “Procedures for Calculating DPA Protection Test Pass/Fail Criteria,” M. Souryal, WINNF-17-I-0289, WG4, Dec. 12, 2017.6. “Revision to Interference Margin Allocation Path Loss Calculation Requirement,” M. Souryal, WINNF-17-I-0278, WG1, Dec.
4, 2017.7. “SAS Functional Tests of Interference Protection with Multiple Constraints,” M. Souryal, A. Sahoo, and T. Nguyen, WINNF-
17-I-00186, WG4, Aug. 2017.8. “SAS Functional Tests of Federal Incumbent Protection,” M. Souryal, WINNF-17-I-00167, WG4, Aug. 2017.9. “Proportional Fair Allocation of Interference Budget,” A. Sahoo, WINNF-17-I-00061, WG1 Coexistence Task Group, Feb.
2017.10. “ESC Requirements,” S. MacMullan and M. Souryal, WINNF-16-I-0263, Rev. 1, WG4 ESC Certification Task Group, Jan. 25,
2017.11. “Methodology for Defining a Protection Zone,” M. Souryal and T. Nguyen, WINNF-16-I-00264, WG4 ESC Certification Task
Group, Dec. 7, 2016.12. “Fairness Considerations for Coexistence of GAA users,” A. Sahoo, WINNF-16-I-0218, WG1 Coexistence Task Group, Oct.
2016. 34